CN-115830265-B - Automatic driving movement obstacle segmentation method based on laser radar

CN115830265BCN 115830265 BCN115830265 BCN 115830265BCN-115830265-B

Abstract

The invention discloses an automatic driving movement obstacle segmentation method based on a laser radar, which comprises the steps of obtaining time sequence point cloud data of a plurality of continuous time stamps in an automatic driving scene, wherein the time sequence point cloud data of the plurality of continuous time stamps comprise time sequence point cloud data of a current time stamp, determining inter-frame difference characteristic information of each time stamp according to the time sequence point cloud data of the plurality of continuous time stamps, inputting the inter-frame difference characteristic information of each time stamp and projection data of the time sequence point cloud data of the current time stamp into a trained segmentation network model, and obtaining a state prediction mask and a semantic prediction mask for obstacle segmentation by using semantic information as guidance. The characteristics of obstacle point cloud data are expressed based on the unique data representation of the point cloud projection and the composition of the end-to-end deep learning network, so that the more accurate identification of the movement obstacle in the automatic driving scene is realized.

Inventors

SUN YUXIANG
MENG SHIYU

Assignees

香港理工大学深圳研究院

Dates

Publication Date: 20260508
Application Date: 20221102

Claims (10)

1. An automatic driving movement obstacle segmentation method based on a laser radar is characterized by comprising the following steps: Acquiring time sequence point cloud data of a plurality of continuous time stamps in an automatic driving scene, wherein the time sequence point cloud data of the plurality of continuous time stamps comprises the time sequence point cloud data of the current time stamp; Determining inter-frame difference characteristic information of each time stamp according to the time sequence point cloud data of a plurality of continuous time stamps; inputting inter-frame difference characteristic information of each time stamp and projection data of time sequence point cloud data of the current time stamp into a trained segmentation network model to obtain a state prediction mask and a semantic prediction mask for obstacle segmentation; the trained segmentation network model comprises a time sequence fusion module, an asymmetric coder and decoder and a point cloud data form recovery module; Inputting the inter-frame difference characteristic information of each time stamp and the projection data of the time sequence point cloud data of the current time stamp into a trained segmentation network model to obtain a state prediction mask and a semantic prediction mask for barrier segmentation, wherein the method comprises the following steps: inputting the inter-frame difference characteristic information of each time stamp and the projection data of the time sequence point cloud data of the current time stamp into the time sequence fusion module to obtain fusion characteristics; Inputting the fusion characteristics into the asymmetric codec to obtain characteristic information data; And inputting the characteristic information data into the point cloud data form recovery module to obtain a state prediction mask and a semantic prediction mask for obstacle segmentation.
2. The laser radar based automatic driving movement obstacle segmentation method according to claim 1, wherein the asymmetric codec comprises an encoder, a context information module, a movement segmentation decoder and a semantic information decoder, wherein the characteristic information data comprises an obstacle movement state prediction mask and a movement obstacle semantic prediction mask; inputting the fusion features into the asymmetric codec to obtain feature information data, including: Inputting the fusion characteristic into the encoder to obtain an encoding characteristic; inputting the coding features into the context information module to obtain feature images; inputting the characteristic image into the motion segmentation decoder to obtain an obstacle motion state prediction mask; and inputting the characteristic image into the semantic information decoder to obtain a semantic prediction mask of the movement obstacle.
3. The laser radar based autopilot movement obstacle segmentation method as set forth in claim 2, wherein the context information module comprises four parallel inflated convolution layers with inflation rates of 6, 12, 18 and 24, respectively; the encoder, the motion segmentation decoder and the semantic information decoder all comprise 4 encoding modules, and each encoding module comprises a convolution layer, a batch normalization layer, a residual error module and an activation function layer.
4. The method for dividing the automatic driving movement obstacle based on the laser radar according to claim 1, wherein the determining the inter-frame difference characteristic information of each time stamp according to the time sequence point cloud data of a plurality of continuous time stamps comprises: Uniformly converting all point cloud coordinate information of past time stamps into a current time stamp point cloud data coordinate system according to pose transformation, and projecting according to time sequence point cloud data of affine transformation continuous time stamps to obtain projection data of the time sequence point cloud data of each time stamp; And carrying out pixel-by-pixel multiplication operation and normalization processing on the projection data of the time sequence point cloud data of each past time stamp in the time sequence point cloud data of a plurality of continuous time stamps according to the projection data of the time sequence point cloud data of the past time stamp and the projection data of the time sequence point cloud data of the current time stamp to obtain the inter-frame difference characteristic information of the past time stamp.
5. The laser radar-based automatic driving movement obstacle segmentation method according to claim 4, wherein the inter-frame difference feature information is: Wherein diff represents inter-frame difference feature information, SP current represents projection data of time-series point cloud data of a current time stamp, and SP i represents projection data of time-series point cloud data of an i-th past time stamp.
6. The laser radar-based automatic driving movement obstacle segmentation method according to claim 1, the method is characterized in that the method for segmenting the autopilot movement obstacle based on the laser radar further comprises the following steps: And determining the dynamic and static categories of the obstacle according to the state prediction mask of the obstacle segmentation, and determining the semantic categories of the obstacle according to the semantic prediction mask of the obstacle segmentation.
7. The laser radar-based automatic driving movement obstacle segmentation method according to claim 1, wherein the trained segmentation network model updates model parameters of the segmentation network model based on an overall loss function during training, and the overall loss function is: Loss=L semantic +L moving +L ls Where Loss represents the overall Loss function, L semantic represents the semantic cross entropy Loss function, L moving represents the movement obstacle Loss function, and L ls represents the lovassz Softmax Loss function.
8. The laser radar-based automatic driving movement obstacle segmentation method according to claim 1, wherein the trained segmentation network model is evaluated based on precision, a cross-over ratio and network thrust time during training, and the cross-over ratio is: Wherein IoU denotes an intersection ratio, target ∈ denotes the number of pixels in a common region between the target mask and the prediction mask, and target ∈ denotes the total number of pixels between the target mask and the prediction mask.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 8 when the computer program is executed.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 8.

Description

Automatic driving movement obstacle segmentation method based on laser radar Technical Field The invention relates to the technical fields of automatic driving and computer vision, in particular to an automatic driving movement obstacle segmentation method based on a laser radar. Background With the development of computer vision technology, cameras are widely used, and can capture very rich environmental information, and can perform region segmentation by utilizing characteristics such as colors, textures and the like, but the cameras are easily interfered by natural conditions such as illumination, weather and the like. The laser radar (LiDAR) has the advantages of long measurement distance, high precision, no influence of weather factors such as shading caused by illumination, good stability and data robustness, so the scheme is based on the research of laser radar data development. Movement disorders mainly include pedestrians, moving vehicles, etc. Under a dynamic traffic environment, movement barriers are ubiquitous and inevitably. The semantic segmentation task is the basis of the point cloud segmentation task of the movement obstacle. The semantic segmentation of autonomous driving is a dense classification task from point to face, and can effectively distinguish categories such as sky, trees, vehicles, bicycles and the like in one piece of scanning data. Semantic segmentation may be considered as a previous step in moving object segmentation. At present, liDAR semantic segmentation tasks can be divided into three main stream methods, namely a point-based, projection-based and voxel-based discrimination method. Among these, more attention is paid to a projection-based research method in which a mature neural network in the field of two-dimensional image segmentation can be directly used. Unlike semantic segmentation, moving object segmentation is not only a point-wise dense mapping task, but also requires efficient extraction of differences between successive scan frames to distinguish dynamic object obstructions. Dynamic and static identification of target objects in an autopilot environment is a key to achieving safe motion planning and navigation. The course of an autonomous car must take into account future coordinates and speeds of surrounding moving objects. Currently, obstacle detection is generally based on two steps, firstly searching for obstacles based on road segmentation and secondly distinguishing whether the obstacles are moving or stationary, and it is difficult to distinguish between background and other moving objects because the vehicle is in a moving state. It is therefore necessary to estimate and compensate for the self-motion of the vehicle in order to extract moving objects in the scene. The traditional segmentation method has certain limitation, such as road segmentation is based on a flat road assumption, has poor generalization capability in potholes and uphill and downhill slopes, and is also influenced by GPS signal intensity in self-motion estimation and compensation. At present, the semantic segmentation method for deep learning is more studied, and can replace the traditional road segmentation method to relieve the assumption problem of a flat road. The obstacle segmentation detection method based on two steps can solve a certain practical application problem, but because training targets in all stages are inconsistent, deviation from an integral macroscopic target is possible, and therefore, the optimal detection accuracy is difficult to achieve finally. Accordingly, the prior art is still in need of improvement and development. Disclosure of Invention The invention aims to solve the technical problems that aiming at the defects in the prior art, an automatic driving movement obstacle segmentation method based on a laser radar is provided, and aims to solve the problems that no mature and perfect end-to-end point cloud movement object obstacle segmentation method and the existing segmentation detection effect on target characteristics are not accurate enough in the prior art. The technical scheme adopted for solving the technical problems is as follows: An automatic driving movement obstacle segmentation method based on a laser radar comprises the following steps: Acquiring time sequence point cloud data of a plurality of continuous time stamps in an automatic driving scene, wherein the time sequence point cloud data of the plurality of continuous time stamps comprises the time sequence point cloud data of the current time stamp; Determining inter-frame difference characteristic information of each time stamp according to the time sequence point cloud data of a plurality of continuous time stamps; inputting inter-frame difference characteristic information of each time stamp and projection data of time sequence point cloud data of the current time stamp into a trained segmentation network model to obtain a state prediction mask and a semantic prediction mask for obst